Language models.
A history with receipts.
When a model was announced isn’t always when it became available. Trace the milestones, see the evidence, and choose the date your research actually needs.
o3
Each milestone keeps its date,
source, precision, and confidence.
One model, several honest dates.
Every figure is computed in your browser from the dataset below.
Counts describe the ledger, not the whole industry.
The announcement is not always the launch
Days from a model’s announcement to the day people could first use it, for models with both dates.
anticipation_days). Open weights count from the weights release; API-only models from API, consumer or platform access. “Before” means the model was usable a day or more before it was announced. Bars show the share within each group.Open weights and API-only models, side by side
Each mark is one model, placed at its first public availability date. Lines join each group’s yearly median.
After launch: can you still use it?
Retirement records per model. A developer shutdown removes an API model for everyone; a host dropping a model removes one copy.
Models in the ledger, by year
Year of first public availability.
Evidence behind the dates
Confidence of every dated event, under the rules in the methodology.
Explore the ledger
From first announcement to final retirement.
Open any model to inspect its full timeline.
| MODEL | DEVELOPER | ANNOUNCED | FIRST AVAILABLE ↓ | VIA | CONFIDENCE | EVENTS | Details |
|---|---|---|---|---|---|---|---|
| Loading the ledger… | |||||||
“First available” is derived from availability events, with documented fallbacks. Confidence refers to the displayed date. How to read the data ↗
A date is only as useful
as its evidence.
The ledger keeps the milestone and its source together, so you can choose, filter, and audit the dates behind your analysis.
Read the full methodology01 One model. Distinct milestones.
Announcement, API access, downloadable weights, consumer rollout, and retirement are separate events. For adoption studies, pick the event that matches the population you study.
The headline date uses the earliest global API availability, weights release, or consumer rollout. If those are missing, it falls back to API preview or free-tier access, then a third-party platform listing. The timeline shows which event supplied it.
02 Confidence you can inspect.
Verified: primary-source curation, independent sources agreeing within two days, or a platform reporting its own event. Inferred: a single source or unresolved differences of 3–30 days. Disputed: stated dates differ by more than 30 days, with all values preserved.
A platform’s own record takes precedence over third-party copies. Repository creation timestamps are bounds and do not verify a launch by themselves. Inspect the source and verifier on each event.
03 Precision stays honest.
A month is displayed as a month, never as an invented day. Year and quarter placeholders do not set first public availability. Timestamp dates use UTC; dates read from a page follow the printed date.
Future retirement dates are schedules, not completed shutdowns. Platform-specific retirements apply only to that host. Changed source dates remain in claim history and the reschedules export.
04 Clear scope. Reproducible results.
The unit is a named trained checkpoint from a lab. Base and instruct checkpoints can be separate models. Packaging variants, ordinary quantizations, API aliases, adapters, and community merges are generally excluded.
The repository publishes the pipeline, schema, cross-catalog names, and validation rules. This website is rebuilt from the repository after every daily update. Scale, compute, and other Epoch AI columns are available in the enriched download.
Made to be checked.
Ready for your research.
Take the data, keep the source links, and credit the work.
The dataset is available under CC BY 4.0.
@misc{llmledger2026,
author = {Ansari, Yasaman},
title = {llm-ledger: a provenance-tracked
dataset of dated events in the lifecycle
of large language models},
year = {2026},
url = {https://github.com/YasamanAnsari/llm-ledger}
}